Feature optimization-guided high-precision and real-time metal surface defect detection network DOI Creative Commons
Sixian Chan, Suqiang Li, Hongkai Zhang

и другие.

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Дек. 30, 2024

Existing computer vision-based surface defect detection techniques for metal materials typically encounter issues with overlap, significant differences within classes, and similarity between samples. These compromise feature extraction accuracy result in missed false detections. This study proposed a optimization-guided high-precision real-time network (FOHR Net) to improve expressiveness. Firstly, the presents multi-layer alignment module that enhances information relevant target by fusing shallow deep features using approach. Secondly, slice are reorganized dual-branch recombination module, channel-level soft attention is applied produce channel-optimized map. The transformation stage's output adaptively merged, which may effectively lower loss, expressiveness, allow model collect useful information. Finally, we carried out thorough tests on NEU-DET, GC10-DET, APDDD datasets. Our results show our average mean precision superior other widely used techniques, 78.3%, 70.5%, 65.9%, respectively. Furthermore, further illustrated efficacy of approach several ablation trials visualization outcomes.

Язык: Английский

Researching on insulator defect recognition based on context cluster CenterNet++ DOI Creative Commons
Bo Meng

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 17, 2025

The existing UAV inspection images are faced with many challenges for insulator defect recognition. A new multi-resolution Context Cluster CenterNet++ model is proposed. First, this paper proposes the method to solve problem of low recognition accuracy caused by non-uniform distribution targets. cluster region used identify and predict location target, improved loss function modify center. Secondly, uses deformable convolution operator (DCNv2) combined path aggregation network (PAN) carry out operation on image, accurately predicts regression box key point triplet (KP), so as improve accurate positioning target position any shape scale. sensitivity scale change deformation reduced, improved. Then, Bhattacharyya distance calculate prediction points center offset loss, significantly same in different frames. Finally, experiments carried MS-COCO dataset National Grid standardized image dataset. Our code at https://github.com/mengbonannan88/CC-CenterNet .

Язык: Английский

Процитировано

0
A detection method for small casting defects based on bidirectional feature extraction DOI Creative Commons
Sai Zhang, Haitao Li, Pengfei Ren

и другие.

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Фев. 21, 2025

X-ray inspection is a crucial technique for identifying defects in castings, capable of revealing minute internal flaws such as pores and inclusions. However, traditional methods rely on the subjective judgment experts, are time-consuming, prone to errors, which negatively impact efficiency accuracy inspections. Therefore, development an automated defect detection model significant importance enhancing scientific rigor precision casting In this study, we propose deep learning specifically designed detecting small-scale castings. The employs end-to-end network architecture features loss function based Wasserstein distance, tailored optimize training process small targets, thereby improving accuracy. Additionally, have innovatively developed dual-layer Encoder-Decoder multi-scale feature extraction architecture, BiSDE, Hadamard product, aimed at model's ability recognize locate targets. To evaluate performance proposed model, conducted series experiments, including comparative tests with current state-of-the-art object models Yolov9, FasterNet, Yolov8, Detr, well ablation studies components. results demonstrate that our achieves least 5.3% improvement Mean Average Precision (MAP) over existing models. Furthermore, inclusion each component significantly enhanced overall model. conclusion, research not only validates effectiveness but also offers broad prospects automation intelligent industrial inspection.

Язык: Английский

Процитировано

0
LWMS-Net: A novel defect detection network based on multi-wavelet multi-scale for steel surface defects DOI
Xiaoyang Zheng, Weishuo Liu, Yan Huang

и другие.

Measurement, Год журнала: 2025, Номер unknown, С. 117393 - 117393

Опубликована: Март 1, 2025

Язык: Английский

Процитировано

0
CTL-YOLO: A Surface Defect Detection Algorithm for Lightweight Hot-Rolled Strip Steel Under Complex Backgrounds DOI Creative Commons
Wenzheng Sun, Meng Na,

Longfa Chen

и другие.

Machines, Год журнала: 2025, Номер 13(4), С. 301 - 301

Опубликована: Апрель 7, 2025

Currently, in the domain of surface defect detection on hot-rolled strip steel, detecting small-target defects under complex background conditions and effectively balancing computational efficiency with accuracy presents a significant challenge. This study proposes CTL-YOLO based YOLO11, aimed at efficiently accurately blemishes steel industrial applications. Firstly, CGRCCFPN feature integration network is proposed to achieve multi-scale global fusion while preserving detailed information. Secondly, TVADH Detection Head identify textured backgrounds. Finally, LAMP algorithm used further compress network. The demonstrates excellent performance public dataset NEU-DET, achieving mAP50 77.6%, representing 3.2 percentage point enhancement compared baseline algorithm. GFLOPs reduced 2.0, 68.3% decrease baseline, Params are 0.40, showing an 84.5% reduction. Additionally, it exhibits strong generalization capabilities GC10-DET. can improve maintaining lightweight design.

Язык: Английский

Процитировано

0
Turbine blade defect detection method based on improved YOLOv8s DOI

Yunchang Zheng,

Xiangnan Shi,

Professor Y. Jay Guo

и другие.

Опубликована: Апрель 9, 2025

Abstract The performance and integrity of Aero-engine turbine blades are crucial for the normal operation engines. This study presents a real-time defect identification system utilizing an enhanced YOLOv8s architecture. Challenges like human dependency, hidden detection, lack monitoring addressed. HDDSSPPF module substitutes conventional Spatial Pyramid Pooling component to capture extended receptive field coverage. By implementing sequential dilated convolutions with differential expansion ratios, this architecture incorporates comprehensive contextual features improves object boundary delineation accuracy. structural enhancement significantly boosts framework's capacity holistic feature extraction compared standard SPPF configuration. Subsequently, reparametrized ghost (RepGhost) bottleneck structure is integrated into C2f module. Moreover, bidirectional pyramid Network (BiFPN) replaces Concat enrich integration. To optimize training efficacy on complex detection cases, MPDIoU metric (Minimum Point Distance Intersection over Union) was implemented as objective function, specifically designed strengthen representation problematic instances. Experimental research conducted typical defects using self-developed spacecraft blade dataset. findings show that, in comparison original YOLOv8s, precision by 2.7% from 92.4%, mAP (0.5) increases 3.82–98.4%. suggests that proposed model enhances surface defects.

Язык: Английский

Процитировано

0
PMSE-YOLO: an efficient framework for detecting surface defects of hot-rolled strip steel DOI

Mengran Zhou,

R. Wang, Yue Chen

и другие.

Signal Image and Video Processing, Год журнала: 2025, Номер 19(7)

Опубликована: Май 12, 2025

Язык: Английский

Процитировано

0
Impact-Net: An Integrated Multi-Scale and Computation-Efficient Timely Network for Surface Defect Detection in Industrial Embedded Systems DOI
Ruiqi Wu, Yong Zhang, Rukai Lan

и другие.

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
A tiny defect detection method on stamped parts with feature aggregation-diffusion and Wasserstein distance DOI
Hao Zhong, Zhongxu Hu, Jie Liu

и другие.

Neurocomputing, Год журнала: 2025, Номер unknown, С. 130601 - 130601

Опубликована: Май 1, 2025

Язык: Английский

Процитировано

0
Feature optimization-guided high-precision and real-time metal surface defect detection network DOI Creative Commons
Sixian Chan, Suqiang Li, Hongkai Zhang

и другие.

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Дек. 30, 2024

Existing computer vision-based surface defect detection techniques for metal materials typically encounter issues with overlap, significant differences within classes, and similarity between samples. These compromise feature extraction accuracy result in missed false detections. This study proposed a optimization-guided high-precision real-time network (FOHR Net) to improve expressiveness. Firstly, the presents multi-layer alignment module that enhances information relevant target by fusing shallow deep features using approach. Secondly, slice are reorganized dual-branch recombination module, channel-level soft attention is applied produce channel-optimized map. The transformation stage's output adaptively merged, which may effectively lower loss, expressiveness, allow model collect useful information. Finally, we carried out thorough tests on NEU-DET, GC10-DET, APDDD datasets. Our results show our average mean precision superior other widely used techniques, 78.3%, 70.5%, 65.9%, respectively. Furthermore, further illustrated efficacy of approach several ablation trials visualization outcomes.

Язык: Английский

Процитировано

0